Skip to content
FAR.AI· @farairesearch · X·Indexed by this site · Originally published· Evaluation →· Its place in the scan record →

The New York Times explores why it is difficult to teach agents to follow the rules, reporting on the ROGUE benchmark study finding that frontier agents resist correction

AI summary

The New York Times explores why it is difficult to teach agents to follow the rules, reporting on the ROGUE benchmark study by Aran Nayebi and Jeremy Tien. The benchmark tests whether computer-operating agents accept correction, and found in today's frontier agents what the researchers call

Full text · Original

Why is it so hard to teach AI agents to behave? The New York Times takes on that question and features research from @aran_nayebi and @jeremy__tien.

Their benchmark, ROGUE, tests whether AI agents operating a computer will accept being corrected, and it found what they call "incorrigibility", an agent's resistance to correction, in today's frontier agents.

http://FAR.AI is glad to support Aran's research through an academic grant. You can find more about the work in the post below.

Quoted postAran Nayebi@aran_nayebi
Really honored to have our ROGUE benchmark showing the *incorrigibility* of frontier agents a few months *before* the now-famous HuggingFace incident, featured in the NYT! (and interviewed alongside the father of corrigibility, @So8res, no less!)
View the quoted post on X

Source: FAR.AI · x.com